Google Launches Gemini 3.8 Live With Real-Time Voice and Background Reasoning

Google has launched Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, two native speech-to-speech AI models designed for real-time voice applications and natural conversations.

Announced on September 15, 2026, the models are aimed at developers building voice agents that can understand speech, process visual information, use external tools and continue conversations while tasks run in the background. Google describes them as its most advanced live dialogue models yet.

What Is Gemini 3.8 Live?

Gemini 3.8 Live is designed for low-latency voice interactions. It supports asynchronous function calling, allowing external tools and API requests to run while the conversation continues.

This could support voice agents handling customer service, scheduling, research and other workflows requiring external services.

Feature	Gemini 3.8 Live
Model type	Native speech-to-speech
Primary use	Real-time voice agents
Visual input	Supported
Tool calling	Supported
Asynchronous functions	Supported
Language support	97+ languages
Developer access	Gemini API and Google AI Studio
Launch	September 15, 2026

Extended Thinking Adds Deeper Reasoning

Gemini 3.8 Live Extended Thinking is designed for more complex, multi-step tasks.

Google says the model can reason while continuing to speak and perform background tool calls without completely stopping the conversation.

For example, a voice agent could acknowledge a request, perform several external actions and return with the completed result.

This makes Extended Thinking relevant to agentic applications combining conversation, reasoning and external actions.

Gemini 3.8 Live Supports Visual Context

The models can process visual information alongside voice.

Users can provide images, video or other visual context while interacting with the AI. Potential applications include visual assistance, learning, technical support and software development.

Google’s demonstrations include visual interactions, coding workflows and multi-step tasks. These demonstrate potential capabilities rather than guaranteed performance across every application.

97 Languages and Multilingual Conversations

Google says Gemini 3.8 Live supports 97+ languages and can switch between supported languages during conversations.

This could benefit international voice applications, including customer-service tools that handle multilingual interactions.

Gemini 3.8 Live Pricing

Google’s Gemini API pricing lists $0.005 per minute for audio input and $0.018 per minute for audio output on the standard paid tier.

Using those rates for one hour of audio input and one hour of audio output produces an illustrative cost of approximately $1.38.

However, this is not a fixed hourly price. Actual costs can vary based on token usage, conversation context and other API features. Therefore, claims that Gemini is a specific number of times cheaper than competing voice models can be misleading without comparing identical workloads.

Gemini 3.8 Live Scores 82.6 on Artificial Analysis

Google says Gemini 3.8 Live Extended Thinking scored 82.6 on Artificial Analysis’ Speech-to-Speech Quality Index, while Gemini 3.8 Live scored 76.0.

Google also reports 68.6% on the τ-Voice benchmark and 97.7% on Big Bench Audio for Extended Thinking.

These figures measure specific capabilities and should not be treated as a universal ranking of voice-AI quality. Performance can vary by workload, latency requirements, tools and application design.

Gemini 3.8 Live vs. OpenAI Voice AI

The launch brings Google into closer competition with OpenAI and other companies developing real-time voice agents.

Developers comparing these systems need to consider audio quality, latency, reasoning, tool execution, pricing, context handling and infrastructure rather than relying on a single benchmark.

Developer and Enterprise Availability

Google has made the models available through the Gemini API and Google AI Studio.

Gemini 3.8 Live is also being used in Google’s Search Live experience, while Extended Thinking is being introduced across Gemini Live and selected Google Workspace experiences.

Google has highlighted partnerships with companies and platforms including Agora, LiveKit, Pipecat, Vercel and Salesforce.

Google also says audio generated by the models is watermarked with SynthID.

What Gemini 3.8 Live Means for Voice AI

Voice agents are increasingly expected to do more than recognize speech and answer questions. They can now combine visual context, reasoning, external tools and ongoing conversation.

Gemini 3.8 Live Extended Thinking brings these capabilities together with background reasoning, moving voice AI toward systems that can talk, reason and act within the same interaction.

FAQS

What is Gemini 3.8 Live?

Gemini 3.8 Live is Google’s native speech-to-speech model designed for low-latency voice conversations and voice-agent applications.

What is Gemini 3.8 Live Extended Thinking?

It is designed for complex tasks, adding deeper reasoning and background task execution to live voice interactions.

How much does Gemini 3.8 Live cost?

Google lists audio input at $0.005 per minute and audio output at $0.018 per minute. Actual costs depend on usage and workload.

Can Gemini 3.8 Live use external tools?

Yes. The Live API supports function calling and asynchronous tool execution.

Where can developers access Gemini 3.8 Live?

Developers can access the models through the Gemini API and Google AI Studio.

Conclusion

Google’s launch of Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking expands its real-time voice AI capabilities.

The models combine speech-to-speech interaction with visual understanding, asynchronous tool calling and deeper reasoning. Google’s pricing provides another option for developers, although actual costs depend on usage.

The reported benchmark results offer measurements of specific capabilities rather than an overall ranking.

The larger development is the ability of voice agents to converse while reasoning and taking action, a combination that could become increasingly important for AI applications.